Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 6

Question 50

A user cancels a long research job while workers are still running. Which design best avoids continued unnecessary work and accidental publication?

  1. Hide the progress indicator and leave workers unchanged
  2. Delete the user's browser session immediately
  3. Propagate cancellation and gate final publication on job state
  4. Tell the coordinator to ignore future worker text

Explanation

Correct Answer

C:

Workers need a cancellation signal, and the publisher must check the authoritative job state. Cancellation is a lifecycle condition, not just a user-interface change.

Incorrect Answers

A:

Removing the indicator changes presentation only. Computation and downstream publication can continue unless execution components receive and honor cancellation.

B:

Session removal does not necessarily terminate server-side jobs. The backend needs an explicit cancellation contract for already accepted work.

D:

Ignoring output may avoid synthesis, but workers still consume resources. It also fails to constrain any independent publication or side-effect path.

Question 51

A document review spans several hours and must survive a worker restart. Which state should be durably stored?

  1. Only the final response once the job finishes
  2. Completed task outputs and the remaining work plan
  3. Only the worker's current memory usage
  4. The original request without intermediate artifacts

Explanation

Correct Answer

B:

A durable checkpoint lets replacement workers resume known work rather than infer progress from a lost process. Stored state should identify versions and completion status.

Incorrect Answers

A:

A final-only record provides no recovery point if the worker stops earlier. The hours of completed intermediate work remain vulnerable to loss.

C:

Resource metrics help operations but do not encode completed analysis or pending tasks. They cannot reconstruct the logical progress of the job.

D:

The request allows a complete restart, but cannot preserve completed effort. That fails the stated requirement to survive interruption efficiently.

Question 52

A coordinator spawns four agents, each allowed to spend the entire parent budget. Total cost exceeds the project ceiling. What should change?

  1. Allocate child budgets from a shared remaining allowance
  2. Increase the parent prompt's emphasis on economy
  3. Measure total spend only after all workers finish
  4. Give each worker a copy of the original ceiling

Explanation

Correct Answer

A:

Subtask limits must compose into the overall constraint. A coordinator or budget service can reserve and account for spending across concurrent workers.

Incorrect Answers

B:

Instructions may encourage restraint but do not change the arithmetic of four independently available budgets. A shared limit needs explicit accounting.

C:

After-the-fact measurement reveals overspending but cannot prevent it. The ceiling must influence admission and continued execution before costs are incurred.

D:

Duplicating the ceiling reproduces the failure. Each worker would still consider the whole allowance available regardless of spending by others.

Question 53

Agents may create subagents that can create more subagents. A small request unexpectedly launches hundreds of tasks. Which control directly bounds this expansion?

  1. Limit only the length of each final answer
  2. Trust each child to decide its own maximum depth
  3. A per-agent concurrency limit with no task-tree limit
  4. Enforce depth, child-count, and total-work limits

Explanation

Correct Answer

D:

Local delegation decisions can multiply across levels. Limits on the task tree and aggregate work bound growth even when each individual delegation seems reasonable.

Incorrect Answers

A:

Output length does not constrain how many workers are created. The explosion can occur before any worker produces its final response.

B:

Independent local limits may not compose into a global bound. The coordinating system must enforce constraints across the complete delegation tree.

C:

Limiting simultaneous children per agent can still permit exponential growth across recursive levels. The complete delegation tree needs aggregate bounds.

Question 54

A worker analyzes a large spreadsheet but the coordinator only needs exceptions and totals. What should the worker return for efficient, verifiable synthesis?

  1. A compact result with references to supporting artifacts
  2. Only a claim that analysis completed successfully
  3. Every intermediate tool response in chronological order
  4. A natural-language summary with all numbers removed

Explanation

Correct Answer

A:

The handoff should preserve actionable findings and a way to inspect evidence. Sending the whole transcript adds irrelevant material to the coordinator's context.

Incorrect Answers

B:

Completion status does not provide the requested findings. The coordinator cannot verify totals or investigate exceptions without results and evidence references.

C:

A complete transcript can be useful for audit storage, but is excessive for routine synthesis. It can bury the exceptions and totals required here.

D:

Removing numeric evidence prevents the required aggregation. Concision must preserve the information the next stage actually needs to use.

Question 55

Parallel analysts disagree about an account's risk classification because they applied different policy versions. What should synthesis do first?

  1. Ask a new analyst to vote without the version metadata
  2. Resolve policy-version applicability before combining conclusions
  3. Average the classifications numerically
  4. Use the newest policy without checking its effective date for this account

Explanation

Correct Answer

B:

The disagreement may arise from incompatible decision rules rather than different evidence. Establishing the applicable version makes the subsequent comparison meaningful.

Incorrect Answers

A:

Removing the cause of disagreement deprives the reviewer of decisive information. Another vote could repeat the same mismatch without recognizing it.

C:

An average across incompatible policies does not represent a valid decision under either one. The policy basis must be reconciled before aggregation.

D:

The newest version need not govern a historical or transitional case. Applicability must be resolved before choosing among conclusions based on different rules.

Question 56

Incoming analysis jobs temporarily exceed worker capacity. The service must protect interactive requests from overload. Which architecture best handles the burst?

  1. Use one unbounded FIFO queue for interactive and background work
  2. Store all pending jobs in each worker's memory
  3. Bound admission and queue asynchronous work with separate capacity
  4. Spawn one worker for every incoming job immediately

Explanation

Correct Answer

C:

Backpressure prevents unbounded accumulation from exhausting the service. Separating workload classes preserves capacity for requests with tighter latency requirements.

Incorrect Answers

A:

An unlimited shared queue can increase interactive delay throughout the burst. It neither bounds accumulation nor reserves capacity for latency-sensitive requests.

B:

Duplicating pending work wastes memory and does not provide admission control. Worker restarts can also lose an uncoordinated in-memory backlog.

D:

Unlimited expansion can exhaust shared quotas, memory, or dependencies. It transfers the burst downstream rather than controlling its effect on interactive traffic.

Question 57

A dependency is failing persistently, and repeated calls keep consuming all request threads. What mechanism best contains the failure while allowing later recovery?

  1. Retry every failure with no overall time limit
  2. Cache the last exception as a permanent business answer
  3. Add request threads while leaving the failing call path unchanged
  4. Temporarily open a circuit and probe recovery under control

Explanation

Correct Answer

D:

A circuit breaker stops repeatedly consuming resources on a failing path. Limited recovery probes permit restoration without immediately recreating the overload.

Incorrect Answers

A:

Unlimited retries keep resources tied to the unavailable dependency. They can prevent healthy requests from being served during the incident.

B:

An exception is an execution failure, not domain evidence. Permanent caching also prevents the system from recognizing that the dependency recovered.

C:

More threads can postpone saturation but continue spending resources on the unavailable dependency. Containment requires limiting calls to that path and controlling recovery probes.

Question 58

A fallback model is faster but cannot process the image inputs used by the primary path. Which failover plan is valid?

  1. Use an evaluated compatible path or an explicit degraded response
  2. Send the same image request and assume feature parity
  3. Treat successful HTTP delivery as successful failover
  4. Drop images silently and present the result as complete

Explanation

Correct Answer

A:

A fallback must satisfy the required input and task contract. If that is impossible, the application should expose the limitation rather than pretend equivalent processing.

Incorrect Answers

B:

Model interchangeability cannot be inferred from both being language models. Unsupported input capabilities can make the replacement path invalid before quality is assessed.

C:

Transport success does not establish that the alternate processed the required inputs. Functional and quality compatibility must be evaluated separately.

D:

Removing task evidence changes the meaning of the output. A confident text-only answer would conceal the unsupported portion of the request.

Question 59

An assistant reaches its investigation limit without resolving a complex case. Staff can continue from its work. What handoff best preserves value?

  1. Send all logs without indicating the unresolved question
  2. Transfer evidence, attempted steps, unresolved issues, and ownership
  3. Return a confident answer to avoid staff workload
  4. Transfer only the original request to the shared queue

Explanation

Correct Answer

B:

A useful escalation makes progress and uncertainty visible so the receiving person can continue efficiently. It should also identify who is responsible for the next action.

Incorrect Answers

A:

Raw logs may support investigation but are difficult to act on alone. The handoff should identify what remains unknown and why escalation occurred.

C:

Unresolved evidence does not justify inventing a conclusion. The stated fallback is human continuation, not concealment of the unresolved case.

D:

This discards completed investigation and forces repeated work. The recipient lacks the context needed to distinguish an untried step from a failed one.

Question 60

A workflow can validate stock, reserve payment, and dispatch goods. Dispatch is irreversible in this system. Which sequence minimizes avoidable compensation?

  1. Let the model choose a new order for each transaction
  2. Dispatch first to minimize visible customer latency
  3. Validate prerequisites before the irreversible dispatch step
  4. Execute every step concurrently regardless of dependencies

Explanation

Correct Answer

C:

Cheap reversible checks should establish feasibility before an action that cannot be undone. This ordering reduces the chance of discovering a preventable failure too late.

Incorrect Answers

A:

Variable sequencing adds risk to a known business dependency. The application can enforce the required order without relying on generated plans.

B:

Early dispatch commits the consequence before prerequisites are established. A later stock or payment problem may then require expensive manual recovery.

D:

Concurrency is appropriate only when results do not depend on one another. Here dispatch must rely on successful prerequisite checks and reservations.

Claude Certified Architect - Professional (CCAR-P) Practice Tests